Science Inventory

Application of Metamodeling Approaches to Improve Cohort-Specific High-Throughput Exposure Predictions

Citation:

Jacketti, M., Z. Stanfield, C. Ring, J. Wambaugh, AND K. Isaacs. Application of Metamodeling Approaches to Improve Cohort-Specific High-Throughput Exposure Predictions. ISES, Chicago, IL, August 28 - 31, 2023. https://doi.org/10.23645/epacomptox.24030021

Impact/Purpose:

This abstract describes a proposed presentation at ISES23 on a new version of the ExpoCast SEEM model

Description:

The U.S. Environmental Protection Agency’s (EPA) Exposure Forecasting (ExpoCast) project has developed high-throughput (HT) models for estimating population exposures to thousands of chemicals for use in chemical prioritization. Within the ExpoCast Systematic Empirical Evaluation of Models (SEEM) framework, predictions from multiple process-based models are integrated with biomonitoring data to produce calibrated consensus predictions for hundreds of thousands of chemicals. However, for chemicals with no specific predictions from any of the component models, SEEM relies on the use of poorly resolved mean values when estimating consensus exposure. In addition, the current version of the SEEM framework does not consider stratification by age or gender cohort. Recently, machine learning-based metamodeling approaches have increasingly been used to model the response of complex exposure models, reducing computational time and parameterization burden thus allowing for the efficient expansion of predictions to new chemicals or model conditions. Here a metamodeling framework was applied to a process-based consumer exposure model, (the HT Stochastic Human Exposure and Dose Simulation model, SHEDS-HT) to expand SHEDS-HT predictions to thousands of additional chemicals based on chemical properties and occurrence in specific product use categories (PUCs). The framework comprised a multilabel classification model that utilized chemical structure descriptors to predict PUC occurrence and a support vector regression model of SHEDS-HT response. Predictions from this framework are being incorporated into a new cohort-specific consensus model for median population exposure. The multilabel classification model and SHEDS-HT metamodel may also be useful in identifying and evaluating reasonably foreseen consumer uses for emerging commercial chemicals. This abstract does not necessarily reflect U.S. EPA policy.